Research map / Detection and segmentation / Open-set and few-shot detection
SAVi++: Towards End-to-End Object-Centric Learning from Real-World Videos
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add depth targets, scaled encoder and augmentation to SAVi (SAVi++) | SAVi · MOVi-E | mIoU (%): 30.7 -> 47.1 | Helped |
| SAVi++ vs SAVi | SAVi · MOVi-D | mIoU (%): 22.7 -> 48.3 | Helped |
| SAVi++ vs SAVi | SAVi · MOVi-A | mIoU (%): 82.3 -> 76.1 | Hurt |
| SAVi++ vs SAVi | SAVi · MOVi-B | mIoU (%): 44.5 -> 25.8 | Hurt |
| SAVi++ vs SAVi | SAVi · MOVi-A | FG-ARI (%): 96.8 -> 98.2 | Mixed |
| remove transformer encoder (ablation) | SAVi++ · MOVi-C/D/E | mIoU | Helped |
| remove data augmentation (ablation) | SAVi++ · MOVi-D/E | mIoU | Helped |
| remove data augmentation (ablation) | SAVi++ · MOVi-C | mIoU | No clear effect |
| remove depth targets (use flow only) (ablation) | SAVi++ · MOVi-E | mIoU | Helped |
| train with depth only, without flow (ablation) | SAVi++ · MOVi-D/E | mIoU | No clear effect |
| add Gaussian noise (10-40cm) to LiDAR depth targets (ablation) | SAVi++ · Waymo Open | B. mIoU: 49.7 -> 50.1 | No clear effect |
| raise input resolution to 256x384 (HR) | SAVi++ · Waymo Open | B. mIoU: 49.7 -> 51.9 | Helped |
Closest papers by meaning
- Conditional Object-Centric Learning from Video (ICLR 2022 · similarity 0.90)
- Segmenting Moving Objects via an Object-Centric Layered Representation (NeurIPS 2022 · similarity 0.87)
- Temporally Consistent Object-Centric Learning by Contrasting Slots (CVPR 2025 · similarity 0.87)
- GLASS: Guided Latent Slot Diffusion for Object-Centric Learning (CVPR 2025 · similarity 0.86)
- Object Concepts Emerge from Motion (NeurIPS 2025 · similarity 0.86)
- DORSal: Diffusion for Object-centric Representations of Scenes $\textit{et al.}$ (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.